Wind power forecasting using LSTM incorporating fourier transformation based denoising technique

Forecasting of wind power is necessary to remove the system operational uncertainties via ensuring more reliable inputs for power system scheduling and control. In order to achieve accurate forecast of wind power for up to 50 seconds (very short-term forecasting), this paper proposes the method Fourier Denoising combined with Long Short-Term Memory (FD-LSTM). The FD-LSTM cascades the output of the Fast Fourier Transform (FFT)-based denoising algorithm to the input of the Long Short-Term Memory (LSTM) forecaster. In this method, first, the inclusion of the FFT-based denoising algorithm ensures better and balanced performance for different forecasting horizons by removing the frequencies. Next, short-term forecasting by employing LSTM reduces the uncertainty and improves both the quality of the operation and the planning. In this paper, we first evaluate the performance of the proposed FD-LSTM method on increased forecasting horizons based on Mean Square Error (MSE), Mean Absolute Error (MAPE), and R-Squared, and compare the results against linear regression. Afterwards, the method is tested with data sets where false data is present. The results show that FDLSTM outperforms the other forecasting methods under the presence of false data.